Vous êtes sur la page 1sur 6



Edward Witten

Notes by Pavel Etingof and David Kazhdan

In this lecture we will consider gauge symmetry breaking.
2.1. Gauge symmetry. Recall what gauge symmetry is. We have a spacetime
X = Rtime  X0. We have a compact gauge group G. We have a eld theory where
a eld con guration is a connection is some principal bundle over Rd and possibly
some matter elds.
Recall the Hamiltonian approach to gauge theory. Let M~ 0 be the space of so-
lutions to the classical equations of motion. On M~ 0 we have an action of the
group G^ of gauge transformations. Let M0  M~ 0 be the space of all solutions
where the G-bundle is trivialized in the time direction, and the connection is trivial
in that direction. Such solutions as usual are completely determined by the pair
A(t0 ); dA
dt (t0 ), where t = t0 is a space cycle, and initial data for the matter elds. It
is clear that any element of M~ 0 can be brought to M0 by a gauge transformation,
so M0 still contains all solutions up to gauge transformations.
Suppose that X0 = Rd 1. In this case we may consider only trivial bundles, and
connections which vanish at spatial in nity. In other words, if A is the space of
connections A on the trivial G-bundle over Rd 1 which vanish at 1 then M0 for
pure gauge theory is T  A. If matter elds are present, then M0 is a product of
T  A with some other space.
De ne G~ to be the group of elements of Maps(Rd 1; G) which have a limit at
in nity, and G~ 0 to be the subgroup of G~ consisting of functions which tend to 1 at
1. We have G= ~ G~ 0 = G. This quotient group is called the group of constant gauge
transformations at 1 and called G1.
The group G~ acts symplectically on M0 . The physical phase space in gauge
theory is the symplectic quotient M = M0 ==G~ 0. Note that we only divide by G~ 0
and not by the whole group G~ , so that we inherit an action of the quotient G1 on
M . It is this symmetry group whose breaking we will discuss.
2.2. Breaking of gauge symmetry and charges at in nity.
De nition. Suppose we have a (classical) gauge theory, and let s 2 M be its
vacuum state. Let H  G = G1 be the stabilizer of s. In this case we will say that
at the vacuum state s the gauge symmetry is broken from G to H .
Thus, by symmetry breaking we mean essentially the same thing as for global
symmetry: there is a symmetry of the Poisson algebra of functions on M which
does not x a particular vacuum state.
Important remark. The above expression \the same thing " should be taken
with great care. There are some fundamental di erences between the two situations,
Typeset by -T X
which will become clear below. They come from the fact that in the situation we
are considering here, (unlike Lecture II1) the physical observables, being gauge
invariant by de nition, automatically commute with G and therefore do not, in
general, separate points on M ; i.e. not every function on M is \observable". In
other words, the action of G on the \theory" (in the sense of Lecture II1) is trivial
from the beginning.
Let us now compute the action of G (classically). First of all, we have a moment
map  : M0 ! g~0 , where g~0 is the Lie algebra of G~ 0 { the algebra of functions from
Rn 1 to the Lie algebra g of G which vanish at in nity. Thus, for any " 2 g~0 we
have a Hamiltonian Q(") 2 C 1(M0 ) de ned by Q(")(X ) = (X )(").
In fact, it is easy to compute Q(") using Noether formalism. Namely,
(2.1) Q(") = Tr ( @A
dt rA ") dd 1 x + matter terms ;
Rd 1

On M , Q(") = 0 if " vanishes at in nity. Thus, on M we have rA dA

dt =

matter terms . In particular, in pure gauge theory rA dt = 0.
Taking this into account, we see that on M
(2.2) Q(") = Tr (rA (" dA d 1
dt ))d x:
Rd 1

Using Stokes' formula, we can rewrite (2.2) as

(2.3) Q(") = rlim dA
d 1Tr(" dt ) = rlim dTr("F );
!1 Sn 2 (r) !1 S n r
2( )

where F is the curvature of the spacetime connection corresponding to the given

point of M0 , and S k (r) is the k-sphere of radius r. This formula de nes the hamil-
tonians for the action of G = G1 on M .
This formula shows that Q(") vanishes for all gauge transformations (not nec-
essarily vanishing at in nity) on a particular state if F = o(r2 n ), r ! 1 on that
state. However, if this is not the case, then Q(") may be nonzero for a constant ".
Example. Consider a U (1) gauge theory with a charged complex scalar. The
elds are a connection A on a hermitian line bundle and a section  of this bundle.
The Lagrangian is
(2.3) L = 4e2 F + jDA j d x + 8 (jj2 + v2 )2d4 x:
1 2 2 4

This is the most general renormalizable Lagrangian in these elds in 4 dimensions.

Here e; ; v are parameters and e2;  are positive while v2 can be positive or negative.
For simplicity we assume rst that v2 6= 0.
This theory is not believed to exist in the UV, but we will regard it as an e ective
theory for some more fundamental theory.
Classically (and quantum mechanically for e2 ;  << 1) we have two cases.
1. v2 > 0; the potential has a single minimum.
2. v2 < 0; the potential has a circle of minima.
Let us consider how in these two cases the theory behaves in the infrared.

-1 -0.5 00 0.5 1

Figure 1. The potential for v2 > 0.

Case 1. v > 0. In this case the minimum of energy is attained when  = 0.


First consider the case when the gauge coupling vanishes: e2 = 0. In this case our
theory is a direct product of a pure (free) abelian gauge theory and the 4 theory.
Therefore, it has a unique vacuum, and the particles which occur at the lower part
of the spectrum are a massless vector, or gauge boson (coming from gauge theory)
and two massive real scalars (coming for 4 theory).
If we turn on small e2 the situation should remain the same. Indeed, certainly
nothing can happen to the massive scalars (the part of the Hilbert space with the
nonzero charge, where these scalars are, has a mass gap, and massiveness is an open
condition); moreover, their masses must be equal since there is a U (1) symmetry
at in nity (the Q(") for constant ") which prohibits the masses to di er. The fact
that Q(") 6= 0 is clear since this is so at e2 = 0, when Q(") represents the U (1)
global symmetry.
Furthermore, the massless vectors cannot become massive. Indeed, recall that
a massless vector means an irreducible representation of SO(3; 1) with p2 = 0 and
spin 1, i.e. the space of sections of a 2-dimensional equivariant vector bundle over
the light cone. This vector bundle cannot be deformed to an equivariant vector
bundle over the hyperboloid, since the stabilizer group SO(3) of a point on the
hyperboloid does not have an irreducible 2-dimensional representation. Thus, the
quantum theory for small coupling will have the same particles { two massive scalars
(the real and imaginary part of ) and a massless vector (the gauge boson).
Remark. The above argument on non-deformability of a massless vector fails
in 3 and 2 dimensions. For example, in 3 dimensions, the massless vector is just the
space of functions over the cone, which can be successfully deformed into the space
of functions over a hyperboloid. This actuallyRhappens when in pure U (1) gauge
theory one introduces a Chern-Simons term c A ^ dA. The theory remains free
but becomes massive, yielding one massive scalar. In the theory we are considering
(for 3 dimensions), this cannot happen dynamically since the Chern-Simons term
is odd under change of orientation, but in other theories this could happen.
In fact, quantum mechanically the operator Q(") (for a suitable normalization
of ") has integer eigenvalues, and thus de nes (in quantum theory) a Z-grading of
the corresponding Hilbert space. In particular, since Q(") 6= 0, there are sectors
of the Hilbert space which cannot be reached from the vacuum by applying local
operators. This shows that we have a fundamental violation of Wightman axioms:
the representation of the operator algebra in the physical Hilbert space is not irre-
ducible. However, the theory still has one vacuum only: the minimal energy in the
sectors with nonzero charge Q(") is positive.

-2 -1 00 1 2

Figure 2. The potential for v2 < 0.

Case 2. v < 0. Let v = b . Then classically we have a minimum of
2 2 2

energy on the circle jj = b. This implies that any nite energy con guration has
the property  = bei0 at in nity, where 0 is a constant. Therefore, by a gauge
transformation which has a nite limit at in nity, we can arrange that  is real and
positive:  = b + w where w is a new real variable. Writing the Lagrangian in terms
of the new variables, we will get something with the following quadratic part:
(2.4) 1
Lquadratic = 4e2 F + d x((dw) + M w ) + d4xb2 A2 :
2 4 2 2 2

It is seen from (2.4) than now all elds are massive. Of course, Lagrangian (2.4) is
not gauge invariant for A, since we have already \spent" the gauge symmetry on
making  real.
Thus, infrared limit of the corresponding quantum theory is trivial for small
values of the couplings. In particular, there are no massless gauge bosons: they
have been \eaten" by the - eld. This situation is called Higgs phenomenon, or
spontaneous breaking of gauge symmetry.
Note that in spite of the presence of a circle of zero energy states, our theory
has only one vacuum. In other words, all points of the circle are regarded as
the same state, on the grounds that they are gauge equivalent to each other and
therefore de ne equivalent realizations (i.e. give the same expectation values of
gauge invariant local operators) This is a fundamental di erence between gauge
and global symmetry breaking. In global symmetry breaking, the points of the
circle represent di erent vacua (realizations) of the theory, since there exist non-
symmetric operators which have di erent expectation values at di erent point of
the circle.
Note also that the operator Q(") doesn't act in the Hilbert space of states, since
classically Q(") generates a group which rotates the circle and permutes the zero
energy states. In particular, in this case local operators act irreducibly in the
Hilbert space, and there are no sectors which cannot be reached from the vacuum.
This is the di erence between case 2 and case 1: in case 1, as you remember, Q(")
acts in the Hilbert space nontrivially and de nes a splitting into sectors.
Remark. If one tries to compute Q(") in Case 2 (when the symmetry is broken)
using formula (2.3), the answer will be zero since the integrand dies rapidly at
in nity.
The particles which are found in the infrared in the situation of Case 2 are,
according to (2.4), a massive vector (A) and a massive scalar (). There is only
one scalar since  is now real. Thus, at the level of representation theory the Higgs
phenomenon arising in Case 2 boils down to a deformation of representations of the
Poincare group: a massless vector plus a massless scalar is deformed to a massive
vector. Recall for comparison that a massless vector separately cannot be deformed
into a massive representation.
Finally, consider the special case v2 = 0. In this case classically we have no
symmetry breaking as for v2 > 0, and the particles are a massless vector and two
massless scalar. However, it is not expected to be the quantum answer, since this
con guration is not stable under perturbations.
Remark. If the Lagrangian we start with is not IR free (say, it is the Lagrangian
of an asymptotically free gauge theory) then the classical analysis we discussed
above does not apply in quantum theory. In this case the infrared behavior of the
theory is dicult to determine. In particular, it could happen that in the infrared
the gauge group of the ultraviolet theory will be replaced with some completely
di erent group, which is not even a subgroup in the original group.
2.3. Symmetry breaking and gauging. In conclusion, let us discuss the
connection between global symmetry and gauge symmetry breaking. Suppose we
have a Lagrangian L of a eld theory (say in 4 dimensions) which has a global U (1)
symmetry. A typical example is when the theory contains some scalar elds j
which are sections of hermitian vector bundles, and U (1) acts by multiplication of
these sections by einj  This U (1) symmetry can be gauged, by introducing a U (1)
gauge eld A and new Lagrangian
(2.5) 1
Lgauged = 4e2 FA2 + LA;
where LA is L in which all derivatives of j are replaced by covariant derivatives.
The statement is that if L is infrared free then for small gauge couplings the
symmetry breaking behavior of the theories de ned by L and Lgauged is usually the
same. Namely, if there is breaking of global symmetry for L then there is breaking
of gauge symmetry for Lgauged and vice versa.
Indeed, let us consider both cases.
Case 1. No global symmetry breaking. In this case classically the minimum of
energy is at j = 0, and thus there is a U (1)-invariant vacuum. In quantum theory,
U (1) acts in the Hilbert space, and there are no massless particles (Goldstone
bosons) corresponding to U (1). In this case, for small gauge coupling e, the matter
part LA of the Lagrangian almost decouples from the gauge part; so classically we
get a massless gauge boson.
To consider the quantum mechanical situation, we assume that there are no
massless particles in the ungauged theory. In this case, the above classical an-
swer is also quantum mechanical for small couplings, by the non-defomability of
representations from massless to massive. However, if massless particles (say Gold-
stone bosons corresponding to other symmetries which are broken) are present, this
answer may not be true.
Case 2. Global symmetry breaking. In this case at the minimum of energy
some of the j is not zero. There is no invariant vacuum, and there is a Goldstone
boson corresponding to this symmetry breaking. In this case, pick a vacuum state
and a component (jn) which is not zero at this vacuum. In the gauged theory,
we can perform a gauge transformation which will make this component real and
positive. This shows that if there are no other massless particles (in particular, no
other broken global symmetries), all elds in the theory will become massive. This
happens classically, due to Higgs mechanism as in Case 2 above, and also quantum
mechanically for small couplings. Thus, we have breaking of gauge symmetry.